UNCERTAINTY ANALYSIS OF HYDROLOGICAL MODELLING IN INDIAN WATERSHEDS: A COMPARATIVE ANALYSIS OF DIFFERENT UNCERTAINTY SOURCES AND REGIONAL VARIANCES

Abstract

Understanding natural processes, particularly the water cycle, is inherently challenging due to their unpredictability and complex nature. This complexity is especially pronounced when employing hydrological models, where simplification introduces various uncertainties. Failing to acknowledge and address these uncertainties can introduce biases into the model outcomes, potentially influencing subsequent decision-making processes. The study focuses on the quantification of various sources of uncertainties for the Himalayan and Peninsular watersheds of the Indian subcontinent. For analysis of input and parameter uncertainties, SWAT+ model is used on the Alaknanda River Basin within the Indian Himalayan Region (IHR), on the three gauging stations. The SWAT+ model is calibrated using the Latin hypercube sampling (LHS) algorithm and achieved satisfactory performance (NSE values of 0.56, 0.79, and 0.61) and sensitivity analysis revealed that two of the four most sensitive parameters are snow-related parameters. Input uncertainty analysis is done using different precipitation products such as IMD, IMDAA, CHIRPS, and ERA5, while Parameter uncertainty quantification is conducted using diverse parameter sets generated through the LHS algorithm. Furthermore, with the application of 47 lumped conceptual models within MARRMoT framework, assessment of model structure uncertainty underscores the varying importance of processes, particularly snow store, soil moisture, and routing store in the study region. The findings reveal that the inclusion of additional store components beyond these important stores in the model leads to a decline in performance, along with an increase in complexity and uncertainties. Parameter uncertainty comes out to be the most significant source of uncertainty for the Himalayan watershed among the three examined uncertainties. Additionally, to further understand the model selection uncertainty in the Indian watersheds, 47 lumped conceptual models in MARRMoT are utilised in 102 watersheds of Peninsular India. Although identifying the optimal model based on watershed characteristics remains challenging, the study reveals that model selection is less critical in tropical wet watersheds where all the models perform relatively better. Conversely, in semi-arid watersheds, the choice of model significantly impacts performance. These insights will help build the reliability of hydrological models for diverse geographical conditions across the Indian subcontinent.